1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_key_cache.h"
5 #include "btree_write_buffer.h"
6 #include "bkey_methods.h"
7 #include "btree_update.h"
13 #include "extent_update.h"
17 #include "subvolume.h"
20 #include <linux/random.h>
22 #include <asm/unaligned.h>
24 #define x(name, ...) #name,
25 const char * const bch2_inode_opts[] = {
30 static const char * const bch2_inode_flag_strs[] = {
36 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
38 static int inode_decode_field(const u8 *in, const u8 *end,
39 u64 out[2], unsigned *out_bits)
41 __be64 be[2] = { 0, 0 };
42 unsigned bytes, shift;
52 * position of highest set bit indicates number of bytes:
53 * shift = number of bits to remove in high byte:
55 shift = 8 - __fls(*in); /* 1 <= shift <= 8 */
56 bytes = byte_table[shift - 1];
61 p = (u8 *) be + 16 - bytes;
63 *p ^= (1 << 8) >> shift;
65 out[0] = be64_to_cpu(be[0]);
66 out[1] = be64_to_cpu(be[1]);
67 *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
72 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
73 const struct bch_inode_unpacked *inode)
75 struct bkey_i_inode_v3 *k = &packed->inode;
76 u8 *out = k->v.fields;
77 u8 *end = (void *) &packed[1];
78 u8 *last_nonzero_field = out;
79 unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
83 bkey_inode_v3_init(&packed->inode.k_i);
84 packed->inode.k.p.offset = inode->bi_inum;
85 packed->inode.v.bi_journal_seq = cpu_to_le64(inode->bi_journal_seq);
86 packed->inode.v.bi_hash_seed = inode->bi_hash_seed;
87 packed->inode.v.bi_flags = cpu_to_le64(inode->bi_flags);
88 packed->inode.v.bi_sectors = cpu_to_le64(inode->bi_sectors);
89 packed->inode.v.bi_size = cpu_to_le64(inode->bi_size);
90 packed->inode.v.bi_version = cpu_to_le64(inode->bi_version);
91 SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
92 SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
95 #define x(_name, _bits) \
99 ret = bch2_varint_encode_fast(out, inode->_name); \
105 last_nonzero_field = out; \
106 last_nonzero_fieldnr = nr_fields; \
114 BCH_INODE_FIELDS_v3()
118 out = last_nonzero_field;
119 nr_fields = last_nonzero_fieldnr;
121 bytes = out - (u8 *) &packed->inode.v;
122 set_bkey_val_bytes(&packed->inode.k, bytes);
123 memset_u64s_tail(&packed->inode.v, 0, bytes);
125 SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
127 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
128 struct bch_inode_unpacked unpacked;
130 ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
132 BUG_ON(unpacked.bi_inum != inode->bi_inum);
133 BUG_ON(unpacked.bi_hash_seed != inode->bi_hash_seed);
134 BUG_ON(unpacked.bi_sectors != inode->bi_sectors);
135 BUG_ON(unpacked.bi_size != inode->bi_size);
136 BUG_ON(unpacked.bi_version != inode->bi_version);
137 BUG_ON(unpacked.bi_mode != inode->bi_mode);
139 #define x(_name, _bits) if (unpacked._name != inode->_name) \
140 panic("unpacked %llu should be %llu", \
141 (u64) unpacked._name, (u64) inode->_name);
142 BCH_INODE_FIELDS_v3()
147 void bch2_inode_pack(struct bkey_inode_buf *packed,
148 const struct bch_inode_unpacked *inode)
150 bch2_inode_pack_inlined(packed, inode);
153 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
154 struct bch_inode_unpacked *unpacked)
156 const u8 *in = inode.v->fields;
157 const u8 *end = bkey_val_end(inode);
159 unsigned fieldnr = 0, field_bits;
162 #define x(_name, _bits) \
163 if (fieldnr++ == INODE_NR_FIELDS(inode.v)) { \
164 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
165 memset((void *) unpacked + offset, 0, \
166 sizeof(*unpacked) - offset); \
170 ret = inode_decode_field(in, end, field, &field_bits); \
174 if (field_bits > sizeof(unpacked->_name) * 8) \
177 unpacked->_name = field[1]; \
180 BCH_INODE_FIELDS_v2()
183 /* XXX: signal if there were more fields than expected? */
187 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
188 const u8 *in, const u8 *end,
191 unsigned fieldnr = 0;
195 #define x(_name, _bits) \
196 if (fieldnr < nr_fields) { \
197 ret = bch2_varint_decode_fast(in, end, &v[0]); \
203 ret = bch2_varint_decode_fast(in, end, &v[1]); \
214 unpacked->_name = v[0]; \
215 if (v[1] || v[0] != unpacked->_name) \
219 BCH_INODE_FIELDS_v2()
222 /* XXX: signal if there were more fields than expected? */
226 static int bch2_inode_unpack_v3(struct bkey_s_c k,
227 struct bch_inode_unpacked *unpacked)
229 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
230 const u8 *in = inode.v->fields;
231 const u8 *end = bkey_val_end(inode);
232 unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
233 unsigned fieldnr = 0;
237 unpacked->bi_inum = inode.k->p.offset;
238 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
239 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
240 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags);
241 unpacked->bi_sectors = le64_to_cpu(inode.v->bi_sectors);
242 unpacked->bi_size = le64_to_cpu(inode.v->bi_size);
243 unpacked->bi_version = le64_to_cpu(inode.v->bi_version);
244 unpacked->bi_mode = INODEv3_MODE(inode.v);
246 #define x(_name, _bits) \
247 if (fieldnr < nr_fields) { \
248 ret = bch2_varint_decode_fast(in, end, &v[0]); \
254 ret = bch2_varint_decode_fast(in, end, &v[1]); \
265 unpacked->_name = v[0]; \
266 if (v[1] || v[0] != unpacked->_name) \
270 BCH_INODE_FIELDS_v3()
273 /* XXX: signal if there were more fields than expected? */
277 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
278 struct bch_inode_unpacked *unpacked)
280 memset(unpacked, 0, sizeof(*unpacked));
283 case KEY_TYPE_inode: {
284 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
286 unpacked->bi_inum = inode.k->p.offset;
287 unpacked->bi_journal_seq= 0;
288 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
289 unpacked->bi_flags = le32_to_cpu(inode.v->bi_flags);
290 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
292 if (INODE_NEW_VARINT(inode.v)) {
293 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
295 INODE_NR_FIELDS(inode.v));
297 return bch2_inode_unpack_v1(inode, unpacked);
301 case KEY_TYPE_inode_v2: {
302 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
304 unpacked->bi_inum = inode.k->p.offset;
305 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
306 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
307 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags);
308 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
310 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
312 INODEv2_NR_FIELDS(inode.v));
319 int bch2_inode_unpack(struct bkey_s_c k,
320 struct bch_inode_unpacked *unpacked)
322 if (likely(k.k->type == KEY_TYPE_inode_v3))
323 return bch2_inode_unpack_v3(k, unpacked);
324 return bch2_inode_unpack_slowpath(k, unpacked);
327 int bch2_inode_peek_nowarn(struct btree_trans *trans,
328 struct btree_iter *iter,
329 struct bch_inode_unpacked *inode,
330 subvol_inum inum, unsigned flags)
336 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
340 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
341 SPOS(0, inum.inum, snapshot),
342 flags|BTREE_ITER_CACHED);
347 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
351 ret = bch2_inode_unpack(k, inode);
357 bch2_trans_iter_exit(trans, iter);
361 int bch2_inode_peek(struct btree_trans *trans,
362 struct btree_iter *iter,
363 struct bch_inode_unpacked *inode,
364 subvol_inum inum, unsigned flags)
366 int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
367 bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
371 int bch2_inode_write_flags(struct btree_trans *trans,
372 struct btree_iter *iter,
373 struct bch_inode_unpacked *inode,
374 enum btree_update_flags flags)
376 struct bkey_inode_buf *inode_p;
378 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
380 return PTR_ERR(inode_p);
382 bch2_inode_pack_inlined(inode_p, inode);
383 inode_p->inode.k.p.snapshot = iter->snapshot;
384 return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags);
387 int __bch2_fsck_write_inode(struct btree_trans *trans,
388 struct bch_inode_unpacked *inode,
391 struct bkey_inode_buf *inode_p =
392 bch2_trans_kmalloc(trans, sizeof(*inode_p));
395 return PTR_ERR(inode_p);
397 bch2_inode_pack(inode_p, inode);
398 inode_p->inode.k.p.snapshot = snapshot;
400 return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
402 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
405 int bch2_fsck_write_inode(struct btree_trans *trans,
406 struct bch_inode_unpacked *inode,
409 int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
410 __bch2_fsck_write_inode(trans, inode, snapshot));
411 bch_err_fn(trans->c, ret);
415 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
417 struct bch_inode_unpacked u;
418 struct bkey_inode_buf *inode_p;
421 if (!bkey_is_inode(&k->k))
422 return ERR_PTR(-ENOENT);
424 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
426 return ERR_CAST(inode_p);
428 ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
432 bch2_inode_pack(inode_p, &u);
433 return &inode_p->inode.k_i;
436 static int __bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k, struct printbuf *err)
438 struct bch_inode_unpacked unpacked;
441 bkey_fsck_err_on(k.k->p.inode, c, err,
442 inode_pos_inode_nonzero,
443 "nonzero k.p.inode");
445 bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX, c, err,
446 inode_pos_blockdev_range,
447 "fs inode in blockdev range");
449 bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked), c, err,
451 "invalid variable length fields");
453 bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1, c, err,
454 inode_checksum_type_invalid,
455 "invalid data checksum type (%u >= %u",
456 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
458 bkey_fsck_err_on(unpacked.bi_compression &&
459 !bch2_compression_opt_valid(unpacked.bi_compression - 1), c, err,
460 inode_compression_type_invalid,
461 "invalid compression opt %u", unpacked.bi_compression - 1);
463 bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) &&
464 unpacked.bi_nlink != 0, c, err,
465 inode_unlinked_but_nlink_nonzero,
466 "flagged as unlinked but bi_nlink != 0");
468 bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode), c, err,
469 inode_subvol_root_but_not_dir,
470 "subvolume root but not a directory");
475 int bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k,
476 enum bkey_invalid_flags flags,
477 struct printbuf *err)
479 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
482 bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
483 inode_str_hash_invalid,
484 "invalid str hash type (%llu >= %u)",
485 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
487 ret = __bch2_inode_invalid(c, k, err);
492 int bch2_inode_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
493 enum bkey_invalid_flags flags,
494 struct printbuf *err)
496 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
499 bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
500 inode_str_hash_invalid,
501 "invalid str hash type (%llu >= %u)",
502 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
504 ret = __bch2_inode_invalid(c, k, err);
509 int bch2_inode_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
510 enum bkey_invalid_flags flags,
511 struct printbuf *err)
513 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
516 bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
517 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k), c, err,
518 inode_v3_fields_start_bad,
519 "invalid fields_start (got %llu, min %u max %zu)",
520 INODEv3_FIELDS_START(inode.v),
521 INODEv3_FIELDS_START_INITIAL,
522 bkey_val_u64s(inode.k));
524 bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
525 inode_str_hash_invalid,
526 "invalid str hash type (%llu >= %u)",
527 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
529 ret = __bch2_inode_invalid(c, k, err);
534 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
535 struct bch_inode_unpacked *inode)
537 printbuf_indent_add(out, 2);
538 prt_printf(out, "mode=%o", inode->bi_mode);
541 prt_str(out, "flags=");
542 prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1));
543 prt_printf(out, " (%x)", inode->bi_flags);
546 prt_printf(out, "journal_seq=%llu", inode->bi_journal_seq);
549 prt_printf(out, "bi_size=%llu", inode->bi_size);
552 prt_printf(out, "bi_sectors=%llu", inode->bi_sectors);
556 prt_printf(out, "bi_version=%llu", inode->bi_version);
558 #define x(_name, _bits) \
559 prt_printf(out, #_name "=%llu", (u64) inode->_name); \
561 BCH_INODE_FIELDS_v3()
563 printbuf_indent_sub(out, 2);
566 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
568 prt_printf(out, "inum: %llu ", inode->bi_inum);
569 __bch2_inode_unpacked_to_text(out, inode);
572 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
574 struct bch_inode_unpacked inode;
576 if (bch2_inode_unpack(k, &inode)) {
577 prt_printf(out, "(unpack error)");
581 __bch2_inode_unpacked_to_text(out, &inode);
584 static inline u64 bkey_inode_flags(struct bkey_s_c k)
588 return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
589 case KEY_TYPE_inode_v2:
590 return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
591 case KEY_TYPE_inode_v3:
592 return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
598 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
600 return bkey_inode_flags(k) & BCH_INODE_unlinked;
603 int bch2_trigger_inode(struct btree_trans *trans,
604 enum btree_id btree_id, unsigned level,
609 s64 nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
611 if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
613 int ret = bch2_replicas_deltas_realloc(trans, 0);
617 trans->fs_usage_deltas->nr_inodes += nr;
620 bool old_deleted = bkey_is_deleted_inode(old);
621 bool new_deleted = bkey_is_deleted_inode(new.s_c);
622 if (old_deleted != new_deleted) {
623 int ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, new.k->p, new_deleted);
629 if ((flags & BTREE_TRIGGER_ATOMIC) && (flags & BTREE_TRIGGER_INSERT)) {
630 BUG_ON(!trans->journal_res.seq);
632 bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq);
635 if (flags & BTREE_TRIGGER_GC) {
636 struct bch_fs *c = trans->c;
638 percpu_down_read(&c->mark_lock);
639 this_cpu_add(c->usage_gc->b.nr_inodes, nr);
640 percpu_up_read(&c->mark_lock);
646 int bch2_inode_generation_invalid(struct bch_fs *c, struct bkey_s_c k,
647 enum bkey_invalid_flags flags,
648 struct printbuf *err)
652 bkey_fsck_err_on(k.k->p.inode, c, err,
653 inode_pos_inode_nonzero,
654 "nonzero k.p.inode");
659 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
662 struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
664 prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
667 void bch2_inode_init_early(struct bch_fs *c,
668 struct bch_inode_unpacked *inode_u)
670 enum bch_str_hash_type str_hash =
671 bch2_str_hash_opt_to_type(c, c->opts.str_hash);
673 memset(inode_u, 0, sizeof(*inode_u));
676 inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
677 get_random_bytes(&inode_u->bi_hash_seed,
678 sizeof(inode_u->bi_hash_seed));
681 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
682 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
683 struct bch_inode_unpacked *parent)
685 inode_u->bi_mode = mode;
686 inode_u->bi_uid = uid;
687 inode_u->bi_gid = gid;
688 inode_u->bi_dev = rdev;
689 inode_u->bi_atime = now;
690 inode_u->bi_mtime = now;
691 inode_u->bi_ctime = now;
692 inode_u->bi_otime = now;
694 if (parent && parent->bi_mode & S_ISGID) {
695 inode_u->bi_gid = parent->bi_gid;
697 inode_u->bi_mode |= S_ISGID;
701 #define x(_name, ...) inode_u->bi_##_name = parent->bi_##_name;
707 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
708 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
709 struct bch_inode_unpacked *parent)
711 bch2_inode_init_early(c, inode_u);
712 bch2_inode_init_late(inode_u, bch2_current_time(c),
713 uid, gid, mode, rdev, parent);
716 static inline u32 bkey_generation(struct bkey_s_c k)
720 case KEY_TYPE_inode_v2:
722 case KEY_TYPE_inode_generation:
723 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
730 * This just finds an empty slot:
732 int bch2_inode_create(struct btree_trans *trans,
733 struct btree_iter *iter,
734 struct bch_inode_unpacked *inode_u,
735 u32 snapshot, u64 cpu)
737 struct bch_fs *c = trans->c;
739 u64 min, max, start, pos, *hint;
741 unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
743 if (c->opts.shard_inode_numbers) {
744 bits -= c->inode_shard_bits;
747 max = (cpu << bits) | ~(ULLONG_MAX << bits);
749 min = max_t(u64, min, BLOCKDEV_INODE_MAX);
750 hint = c->unused_inode_hints + cpu;
752 min = BLOCKDEV_INODE_MAX;
753 max = ~(ULLONG_MAX << bits);
754 hint = c->unused_inode_hints;
757 start = READ_ONCE(*hint);
759 if (start >= max || start < min)
763 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
764 BTREE_ITER_ALL_SNAPSHOTS|
767 while ((k = bch2_btree_iter_peek(iter)).k &&
768 !(ret = bkey_err(k)) &&
769 bkey_lt(k.k->p, POS(0, max))) {
770 if (pos < iter->pos.offset)
774 * We don't need to iterate over keys in every snapshot once
775 * we've found just one:
777 pos = iter->pos.offset + 1;
778 bch2_btree_iter_set_pos(iter, POS(0, pos));
781 if (!ret && pos < max)
784 if (!ret && start == min)
785 ret = -BCH_ERR_ENOSPC_inode_create;
788 bch2_trans_iter_exit(trans, iter);
792 /* Retry from start */
794 bch2_btree_iter_set_pos(iter, POS(0, pos));
797 bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
798 k = bch2_btree_iter_peek_slot(iter);
801 bch2_trans_iter_exit(trans, iter);
805 *hint = k.k->p.offset;
806 inode_u->bi_inum = k.k->p.offset;
807 inode_u->bi_generation = bkey_generation(k);
811 static int bch2_inode_delete_keys(struct btree_trans *trans,
812 subvol_inum inum, enum btree_id id)
814 struct btree_iter iter;
816 struct bkey_i delete;
817 struct bpos end = POS(inum.inum, U64_MAX);
822 * We're never going to be deleting partial extents, no need to use an
825 bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
829 bch2_trans_begin(trans);
831 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
835 bch2_btree_iter_set_snapshot(&iter, snapshot);
837 k = bch2_btree_iter_peek_upto(&iter, end);
845 bkey_init(&delete.k);
846 delete.k.p = iter.pos;
848 if (iter.flags & BTREE_ITER_IS_EXTENTS)
849 bch2_key_resize(&delete.k,
850 bpos_min(end, k.k->p).offset -
853 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
854 bch2_trans_commit(trans, NULL, NULL,
855 BCH_TRANS_COMMIT_no_enospc);
857 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
861 bch2_trans_iter_exit(trans, &iter);
865 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
867 struct btree_trans *trans = bch2_trans_get(c);
868 struct btree_iter iter = { NULL };
869 struct bkey_i_inode_generation delete;
870 struct bch_inode_unpacked inode_u;
876 * If this was a directory, there shouldn't be any real dirents left -
877 * but there could be whiteouts (from hash collisions) that we should
880 * XXX: the dirent could ideally would delete whiteouts when they're no
883 ret = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
884 bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
885 bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
889 bch2_trans_begin(trans);
891 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
895 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
896 SPOS(0, inum.inum, snapshot),
897 BTREE_ITER_INTENT|BTREE_ITER_CACHED);
902 if (!bkey_is_inode(k.k)) {
903 bch2_fs_inconsistent(c,
904 "inode %llu:%u not found when deleting",
905 inum.inum, snapshot);
910 bch2_inode_unpack(k, &inode_u);
912 bkey_inode_generation_init(&delete.k_i);
913 delete.k.p = iter.pos;
914 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
916 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
917 bch2_trans_commit(trans, NULL, NULL,
918 BCH_TRANS_COMMIT_no_enospc);
920 bch2_trans_iter_exit(trans, &iter);
921 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
924 bch2_trans_put(trans);
928 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
930 struct bch_inode_unpacked *inode)
932 struct btree_iter iter;
935 ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
937 bch2_trans_iter_exit(trans, &iter);
941 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
943 struct bch_inode_unpacked *inode)
945 struct btree_iter iter;
948 ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
950 bch2_trans_iter_exit(trans, &iter);
954 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
955 struct bch_inode_unpacked *inode)
957 return bch2_trans_do(c, NULL, NULL, 0,
958 bch2_inode_find_by_inum_trans(trans, inum, inode));
961 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
963 if (bi->bi_flags & BCH_INODE_unlinked)
964 bi->bi_flags &= ~BCH_INODE_unlinked;
966 if (bi->bi_nlink == U32_MAX)
975 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
977 if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) {
978 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
983 if (bi->bi_flags & BCH_INODE_unlinked) {
984 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
991 bi->bi_flags |= BCH_INODE_unlinked;
994 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
996 struct bch_opts ret = { 0 };
997 #define x(_name, _bits) \
998 if (inode->bi_##_name) \
999 opt_set(ret, _name, inode->bi_##_name - 1);
1005 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
1006 struct bch_inode_unpacked *inode)
1008 #define x(_name, _bits) opts->_name = inode_opt_get(c, inode, _name);
1013 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
1016 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
1018 struct bch_inode_unpacked inode;
1019 int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
1024 bch2_inode_opts_get(opts, trans->c, &inode);
1028 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
1030 struct bch_fs *c = trans->c;
1031 struct btree_iter iter = { NULL };
1032 struct bkey_i_inode_generation delete;
1033 struct bch_inode_unpacked inode_u;
1038 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1039 SPOS(inum, 0, snapshot),
1040 SPOS(inum, U64_MAX, snapshot),
1042 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1043 SPOS(inum, 0, snapshot),
1044 SPOS(inum, U64_MAX, snapshot),
1046 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1047 SPOS(inum, 0, snapshot),
1048 SPOS(inum, U64_MAX, snapshot),
1050 } while (ret == -BCH_ERR_transaction_restart_nested);
1054 bch2_trans_begin(trans);
1056 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1057 SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
1062 if (!bkey_is_inode(k.k)) {
1063 bch2_fs_inconsistent(c,
1064 "inode %llu:%u not found when deleting",
1070 bch2_inode_unpack(k, &inode_u);
1072 /* Subvolume root? */
1073 if (inode_u.bi_subvol)
1074 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1076 bkey_inode_generation_init(&delete.k_i);
1077 delete.k.p = iter.pos;
1078 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1080 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1081 bch2_trans_commit(trans, NULL, NULL,
1082 BCH_TRANS_COMMIT_no_enospc);
1084 bch2_trans_iter_exit(trans, &iter);
1085 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1088 return ret ?: -BCH_ERR_transaction_restart_nested;
1091 static int may_delete_deleted_inode(struct btree_trans *trans,
1092 struct btree_iter *iter,
1094 bool *need_another_pass)
1096 struct bch_fs *c = trans->c;
1097 struct btree_iter inode_iter;
1099 struct bch_inode_unpacked inode;
1102 k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
1107 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1108 if (fsck_err_on(!bkey_is_inode(k.k), c,
1109 deleted_inode_missing,
1110 "nonexistent inode %llu:%u in deleted_inodes btree",
1111 pos.offset, pos.snapshot))
1114 ret = bch2_inode_unpack(k, &inode);
1118 if (S_ISDIR(inode.bi_mode)) {
1119 ret = bch2_empty_dir_snapshot(trans, pos.offset, pos.snapshot);
1120 if (fsck_err_on(ret == -ENOTEMPTY, c, deleted_inode_is_dir,
1121 "non empty directory %llu:%u in deleted_inodes btree",
1122 pos.offset, pos.snapshot))
1128 if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked), c,
1129 deleted_inode_not_unlinked,
1130 "non-deleted inode %llu:%u in deleted_inodes btree",
1131 pos.offset, pos.snapshot))
1136 deleted_inode_but_clean,
1137 "filesystem marked as clean but have deleted inode %llu:%u",
1138 pos.offset, pos.snapshot)) {
1143 if (bch2_snapshot_is_internal_node(c, pos.snapshot)) {
1144 struct bpos new_min_pos;
1146 ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos);
1150 inode.bi_flags &= ~BCH_INODE_unlinked;
1152 ret = bch2_inode_write_flags(trans, &inode_iter, &inode,
1153 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1154 bch_err_msg(c, ret, "clearing inode unlinked flag");
1159 * We'll need another write buffer flush to pick up the new
1160 * unlinked inodes in the snapshot leaves:
1162 *need_another_pass = true;
1169 bch2_trans_iter_exit(trans, &inode_iter);
1172 ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
1176 int bch2_delete_dead_inodes(struct bch_fs *c)
1178 struct btree_trans *trans = bch2_trans_get(c);
1179 bool need_another_pass;
1182 need_another_pass = false;
1185 * Weird transaction restart handling here because on successful delete,
1186 * bch2_inode_rm_snapshot() will return a nested transaction restart,
1187 * but we can't retry because the btree write buffer won't have been
1188 * flushed and we'd spin:
1190 ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1191 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1192 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
1193 ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass);
1195 bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot);
1197 ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1199 * We don't want to loop here: a transaction restart
1200 * error here means we handled a transaction restart and
1201 * we're actually done, but if we loop we'll retry the
1202 * same key because the write buffer hasn't been flushed
1205 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1214 if (!ret && need_another_pass) {
1215 ret = bch2_btree_write_buffer_flush_sync(trans);
1221 bch2_trans_put(trans);